mm/shmem.c

   1 /*
   2  * Resizable virtual memory filesystem for Linux.
   3  *
   4  * Copyright (C) 2000 Linus Torvalds.
   5  *               2000 Transmeta Corp.
   6  *               2000-2001 Christoph Rohland
   7  *               2000-2001 SAP AG
   8  *               2002 Red Hat Inc.
   9  * Copyright (C) 2002-2005 Hugh Dickins.
  10  * Copyright (C) 2002-2005 VERITAS Software Corporation.
  11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  12  *
  13  * Extended attribute support for tmpfs:
  14  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  15  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  16  *
  17  * This file is released under the GPL.
  18  */
  19
  20 /*
  21  * This virtual memory filesystem is heavily based on the ramfs. It
  22  * extends ramfs by the ability to use swap and honor resource limits
  23  * which makes it a completely usable filesystem.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/init.h>
  28 #include <linux/fs.h>
  29 #include <linux/xattr.h>
  30 #include <linux/generic_acl.h>
  31 #include <linux/mm.h>
  32 #include <linux/mman.h>
  33 #include <linux/file.h>
  34 #include <linux/swap.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/string.h>
  37 #include <linux/slab.h>
  38 #include <linux/backing-dev.h>
  39 #include <linux/shmem_fs.h>
  40 #include <linux/mount.h>
  41 #include <linux/writeback.h>
  42 #include <linux/vfs.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/security.h>
  45 #include <linux/swapops.h>
  46 #include <linux/mempolicy.h>
  47 #include <linux/namei.h>
  48 #include <linux/ctype.h>
  49 #include <linux/migrate.h>
  50 #include <linux/highmem.h>
  51
  52 #include <asm/uaccess.h>
  53 #include <asm/div64.h>
  54 #include <asm/pgtable.h>
  55
  56 /* This magic number is used in glibc for posix shared memory */
  57 #define TMPFS_MAGIC     0x01021994
  58
  59 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  60 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  61 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  62
  63 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  64 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
  65
  66 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  67
  68 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  69 #define SHMEM_PAGEIN     VM_READ
  70 #define SHMEM_TRUNCATE   VM_WRITE
  71
  72 /* Definition to limit shmem_truncate's steps between cond_rescheds */
  73 #define LATENCY_LIMIT    64
  74
  75 /* Pretend that each entry is of this size in directory's i_size */
  76 #define BOGO_DIRENT_SIZE 20
  77
  78 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
  79 enum sgp_type {
  80         SGP_QUICK,      /* don't try more than file page cache lookup */
  81         SGP_READ,       /* don't exceed i_size, don't allocate page */
  82         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  83         SGP_WRITE,      /* may exceed i_size, may allocate page */
  84 };
  85
  86 static int shmem_getpage(struct inode *inode, unsigned long idx,
  87                          struct page **pagep, enum sgp_type sgp, int *type);
  88
  89 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
  90 {
  91         /*
  92          * The above definition of ENTRIES_PER_PAGE, and the use of
  93          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
  94          * might be reconsidered if it ever diverges from PAGE_SIZE.
  95          */
  96         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  97 }
  98
  99 static inline void shmem_dir_free(struct page *page)
 100 {
 101         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
 102 }
 103
 104 static struct page **shmem_dir_map(struct page *page)
 105 {
 106         return (struct page **)kmap_atomic(page, KM_USER0);
 107 }
 108
 109 static inline void shmem_dir_unmap(struct page **dir)
 110 {
 111         kunmap_atomic(dir, KM_USER0);
 112 }
 113
 114 static swp_entry_t *shmem_swp_map(struct page *page)
 115 {
 116         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 117 }
 118
 119 static inline void shmem_swp_balance_unmap(void)
 120 {
 121         /*
 122          * When passing a pointer to an i_direct entry, to code which
 123          * also handles indirect entries and so will shmem_swp_unmap,
 124          * we must arrange for the preempt count to remain in balance.
 125          * What kmap_atomic of a lowmem page does depends on config
 126          * and architecture, so pretend to kmap_atomic some lowmem page.
 127          */
 128         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 129 }
 130
 131 static inline void shmem_swp_unmap(swp_entry_t *entry)
 132 {
 133         kunmap_atomic(entry, KM_USER1);
 134 }
 135
 136 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 137 {
 138         return sb->s_fs_info;
 139 }
 140
 141 /*
 142  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 143  * for shared memory and for shared anonymous (/dev/zero) mappings
 144  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 145  * consistent with the pre-accounting of private mappings ...
 146  */
 147 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 148 {
 149         return (flags & VM_ACCOUNT)?
 150                 security_vm_enough_memory(VM_ACCT(size)): 0;
 151 }
 152
 153 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 154 {
 155         if (flags & VM_ACCOUNT)
 156                 vm_unacct_memory(VM_ACCT(size));
 157 }
 158
 159 /*
 160  * ... whereas tmpfs objects are accounted incrementally as
 161  * pages are allocated, in order to allow huge sparse files.
 162  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 163  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 164  */
 165 static inline int shmem_acct_block(unsigned long flags)
 166 {
 167         return (flags & VM_ACCOUNT)?
 168                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
 169 }
 170
 171 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 172 {
 173         if (!(flags & VM_ACCOUNT))
 174                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 175 }
 176
 177 static struct super_operations shmem_ops;
 178 static const struct address_space_operations shmem_aops;
 179 static struct file_operations shmem_file_operations;
 180 static struct inode_operations shmem_inode_operations;
 181 static struct inode_operations shmem_dir_inode_operations;
 182 static struct inode_operations shmem_special_inode_operations;
 183 static struct vm_operations_struct shmem_vm_ops;
 184
 185 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 186         .ra_pages       = 0,    /* No readahead */
 187         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 188         .unplug_io_fn   = default_unplug_io_fn,
 189 };
 190
 191 static LIST_HEAD(shmem_swaplist);
 192 static DEFINE_SPINLOCK(shmem_swaplist_lock);
 193
 194 static void shmem_free_blocks(struct inode *inode, long pages)
 195 {
 196         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 197         if (sbinfo->max_blocks) {
 198                 spin_lock(&sbinfo->stat_lock);
 199                 sbinfo->free_blocks += pages;
 200                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
 201                 spin_unlock(&sbinfo->stat_lock);
 202         }
 203 }
 204
 205 /*
 206  * shmem_recalc_inode - recalculate the size of an inode
 207  *
 208  * @inode: inode to recalc
 209  *
 210  * We have to calculate the free blocks since the mm can drop
 211  * undirtied hole pages behind our back.
 212  *
 213  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 214  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 215  *
 216  * It has to be called with the spinlock held.
 217  */
 218 static void shmem_recalc_inode(struct inode *inode)
 219 {
 220         struct shmem_inode_info *info = SHMEM_I(inode);
 221         long freed;
 222
 223         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 224         if (freed > 0) {
 225                 info->alloced -= freed;
 226                 shmem_unacct_blocks(info->flags, freed);
 227                 shmem_free_blocks(inode, freed);
 228         }
 229 }
 230
 231 /*
 232  * shmem_swp_entry - find the swap vector position in the info structure
 233  *
 234  * @info:  info structure for the inode
 235  * @index: index of the page to find
 236  * @page:  optional page to add to the structure. Has to be preset to
 237  *         all zeros
 238  *
 239  * If there is no space allocated yet it will return NULL when
 240  * page is NULL, else it will use the page for the needed block,
 241  * setting it to NULL on return to indicate that it has been used.
 242  *
 243  * The swap vector is organized the following way:
 244  *
 245  * There are SHMEM_NR_DIRECT entries directly stored in the
 246  * shmem_inode_info structure. So small files do not need an addional
 247  * allocation.
 248  *
 249  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 250  * i_indirect which points to a page which holds in the first half
 251  * doubly indirect blocks, in the second half triple indirect blocks:
 252  *
 253  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 254  * following layout (for SHMEM_NR_DIRECT == 16):
 255  *
 256  * i_indirect -> dir --> 16-19
 257  *            |      +-> 20-23
 258  *            |
 259  *            +-->dir2 --> 24-27
 260  *            |        +-> 28-31
 261  *            |        +-> 32-35
 262  *            |        +-> 36-39
 263  *            |
 264  *            +-->dir3 --> 40-43
 265  *                     +-> 44-47
 266  *                     +-> 48-51
 267  *                     +-> 52-55
 268  */
 269 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 270 {
 271         unsigned long offset;
 272         struct page **dir;
 273         struct page *subdir;
 274
 275         if (index < SHMEM_NR_DIRECT) {
 276                 shmem_swp_balance_unmap();
 277                 return info->i_direct+index;
 278         }
 279         if (!info->i_indirect) {
 280                 if (page) {
 281                         info->i_indirect = *page;
 282                         *page = NULL;
 283                 }
 284                 return NULL;                    /* need another page */
 285         }
 286
 287         index -= SHMEM_NR_DIRECT;
 288         offset = index % ENTRIES_PER_PAGE;
 289         index /= ENTRIES_PER_PAGE;
 290         dir = shmem_dir_map(info->i_indirect);
 291
 292         if (index >= ENTRIES_PER_PAGE/2) {
 293                 index -= ENTRIES_PER_PAGE/2;
 294                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 295                 index %= ENTRIES_PER_PAGE;
 296                 subdir = *dir;
 297                 if (!subdir) {
 298                         if (page) {
 299                                 *dir = *page;
 300                                 *page = NULL;
 301                         }
 302                         shmem_dir_unmap(dir);
 303                         return NULL;            /* need another page */
 304                 }
 305                 shmem_dir_unmap(dir);
 306                 dir = shmem_dir_map(subdir);
 307         }
 308
 309         dir += index;
 310         subdir = *dir;
 311         if (!subdir) {
 312                 if (!page || !(subdir = *page)) {
 313                         shmem_dir_unmap(dir);
 314                         return NULL;            /* need a page */
 315                 }
 316                 *dir = subdir;
 317                 *page = NULL;
 318         }
 319         shmem_dir_unmap(dir);
 320         return shmem_swp_map(subdir) + offset;
 321 }
 322
 323 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 324 {
 325         long incdec = value? 1: -1;
 326
 327         entry->val = value;
 328         info->swapped += incdec;
 329         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
 330                 struct page *page = kmap_atomic_to_page(entry);
 331                 set_page_private(page, page_private(page) + incdec);
 332         }
 333 }
 334
 335 /*
 336  * shmem_swp_alloc - get the position of the swap entry for the page.
 337  *                   If it does not exist allocate the entry.
 338  *
 339  * @info:       info structure for the inode
 340  * @index:      index of the page to find
 341  * @sgp:        check and recheck i_size? skip allocation?
 342  */
 343 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 344 {
 345         struct inode *inode = &info->vfs_inode;
 346         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 347         struct page *page = NULL;
 348         swp_entry_t *entry;
 349
 350         if (sgp != SGP_WRITE &&
 351             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 352                 return ERR_PTR(-EINVAL);
 353
 354         while (!(entry = shmem_swp_entry(info, index, &page))) {
 355                 if (sgp == SGP_READ)
 356                         return shmem_swp_map(ZERO_PAGE(0));
 357                 /*
 358                  * Test free_blocks against 1 not 0, since we have 1 data
 359                  * page (and perhaps indirect index pages) yet to allocate:
 360                  * a waste to allocate index if we cannot allocate data.
 361                  */
 362                 if (sbinfo->max_blocks) {
 363                         spin_lock(&sbinfo->stat_lock);
 364                         if (sbinfo->free_blocks <= 1) {
 365                                 spin_unlock(&sbinfo->stat_lock);
 366                                 return ERR_PTR(-ENOSPC);
 367                         }
 368                         sbinfo->free_blocks--;
 369                         inode->i_blocks += BLOCKS_PER_PAGE;
 370                         spin_unlock(&sbinfo->stat_lock);
 371                 }
 372
 373                 spin_unlock(&info->lock);
 374                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
 375                 if (page)
 376                         set_page_private(page, 0);
 377                 spin_lock(&info->lock);
 378
 379                 if (!page) {
 380                         shmem_free_blocks(inode, 1);
 381                         return ERR_PTR(-ENOMEM);
 382                 }
 383                 if (sgp != SGP_WRITE &&
 384                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 385                         entry = ERR_PTR(-EINVAL);
 386                         break;
 387                 }
 388                 if (info->next_index <= index)
 389                         info->next_index = index + 1;
 390         }
 391         if (page) {
 392                 /* another task gave its page, or truncated the file */
 393                 shmem_free_blocks(inode, 1);
 394                 shmem_dir_free(page);
 395         }
 396         if (info->next_index <= index && !IS_ERR(entry))
 397                 info->next_index = index + 1;
 398         return entry;
 399 }
 400
 401 /*
 402  * shmem_free_swp - free some swap entries in a directory
 403  *
 404  * @dir:   pointer to the directory
 405  * @edir:  pointer after last entry of the directory
 406  */
 407 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
 408 {
 409         swp_entry_t *ptr;
 410         int freed = 0;
 411
 412         for (ptr = dir; ptr < edir; ptr++) {
 413                 if (ptr->val) {
 414                         free_swap_and_cache(*ptr);
 415                         *ptr = (swp_entry_t){0};
 416                         freed++;
 417                 }
 418         }
 419         return freed;
 420 }
 421
 422 static int shmem_map_and_free_swp(struct page *subdir,
 423                 int offset, int limit, struct page ***dir)
 424 {
 425         swp_entry_t *ptr;
 426         int freed = 0;
 427
 428         ptr = shmem_swp_map(subdir);
 429         for (; offset < limit; offset += LATENCY_LIMIT) {
 430                 int size = limit - offset;
 431                 if (size > LATENCY_LIMIT)
 432                         size = LATENCY_LIMIT;
 433                 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
 434                 if (need_resched()) {
 435                         shmem_swp_unmap(ptr);
 436                         if (*dir) {
 437                                 shmem_dir_unmap(*dir);
 438                                 *dir = NULL;
 439                         }
 440                         cond_resched();
 441                         ptr = shmem_swp_map(subdir);
 442                 }
 443         }
 444         shmem_swp_unmap(ptr);
 445         return freed;
 446 }
 447
 448 static void shmem_free_pages(struct list_head *next)
 449 {
 450         struct page *page;
 451         int freed = 0;
 452
 453         do {
 454                 page = container_of(next, struct page, lru);
 455                 next = next->next;
 456                 shmem_dir_free(page);
 457                 freed++;
 458                 if (freed >= LATENCY_LIMIT) {
 459                         cond_resched();
 460                         freed = 0;
 461                 }
 462         } while (next);
 463 }
 464
 465 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
 466 {
 467         struct shmem_inode_info *info = SHMEM_I(inode);
 468         unsigned long idx;
 469         unsigned long size;
 470         unsigned long limit;
 471         unsigned long stage;
 472         unsigned long diroff;
 473         struct page **dir;
 474         struct page *topdir;
 475         struct page *middir;
 476         struct page *subdir;
 477         swp_entry_t *ptr;
 478         LIST_HEAD(pages_to_free);
 479         long nr_pages_to_free = 0;
 480         long nr_swaps_freed = 0;
 481         int offset;
 482         int freed;
 483         int punch_hole = 0;
 484
 485         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 486         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 487         if (idx >= info->next_index)
 488                 return;
 489
 490         spin_lock(&info->lock);
 491         info->flags |= SHMEM_TRUNCATE;
 492         if (likely(end == (loff_t) -1)) {
 493                 limit = info->next_index;
 494                 info->next_index = idx;
 495         } else {
 496                 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 497                 if (limit > info->next_index)
 498                         limit = info->next_index;
 499                 punch_hole = 1;
 500         }
 501
 502         topdir = info->i_indirect;
 503         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
 504                 info->i_indirect = NULL;
 505                 nr_pages_to_free++;
 506                 list_add(&topdir->lru, &pages_to_free);
 507         }
 508         spin_unlock(&info->lock);
 509
 510         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 511                 ptr = info->i_direct;
 512                 size = limit;
 513                 if (size > SHMEM_NR_DIRECT)
 514                         size = SHMEM_NR_DIRECT;
 515                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
 516         }
 517         if (!topdir)
 518                 goto done2;
 519
 520         BUG_ON(limit <= SHMEM_NR_DIRECT);
 521         limit -= SHMEM_NR_DIRECT;
 522         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 523         offset = idx % ENTRIES_PER_PAGE;
 524         idx -= offset;
 525
 526         dir = shmem_dir_map(topdir);
 527         stage = ENTRIES_PER_PAGEPAGE/2;
 528         if (idx < ENTRIES_PER_PAGEPAGE/2) {
 529                 middir = topdir;
 530                 diroff = idx/ENTRIES_PER_PAGE;
 531         } else {
 532                 dir += ENTRIES_PER_PAGE/2;
 533                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 534                 while (stage <= idx)
 535                         stage += ENTRIES_PER_PAGEPAGE;
 536                 middir = *dir;
 537                 if (*dir) {
 538                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 539                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 540                         if (!diroff && !offset) {
 541                                 *dir = NULL;
 542                                 nr_pages_to_free++;
 543                                 list_add(&middir->lru, &pages_to_free);
 544                         }
 545                         shmem_dir_unmap(dir);
 546                         dir = shmem_dir_map(middir);
 547                 } else {
 548                         diroff = 0;
 549                         offset = 0;
 550                         idx = stage;
 551                 }
 552         }
 553
 554         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
 555                 if (unlikely(idx == stage)) {
 556                         shmem_dir_unmap(dir);
 557                         dir = shmem_dir_map(topdir) +
 558                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 559                         while (!*dir) {
 560                                 dir++;
 561                                 idx += ENTRIES_PER_PAGEPAGE;
 562                                 if (idx >= limit)
 563                                         goto done1;
 564                         }
 565                         stage = idx + ENTRIES_PER_PAGEPAGE;
 566                         middir = *dir;
 567                         *dir = NULL;
 568                         nr_pages_to_free++;
 569                         list_add(&middir->lru, &pages_to_free);
 570                         shmem_dir_unmap(dir);
 571                         cond_resched();
 572                         dir = shmem_dir_map(middir);
 573                         diroff = 0;
 574                 }
 575                 subdir = dir[diroff];
 576                 if (subdir && page_private(subdir)) {
 577                         size = limit - idx;
 578                         if (size > ENTRIES_PER_PAGE)
 579                                 size = ENTRIES_PER_PAGE;
 580                         freed = shmem_map_and_free_swp(subdir,
 581                                                 offset, size, &dir);
 582                         if (!dir)
 583                                 dir = shmem_dir_map(middir);
 584                         nr_swaps_freed += freed;
 585                         if (offset)
 586                                 spin_lock(&info->lock);
 587                         set_page_private(subdir, page_private(subdir) - freed);
 588                         if (offset)
 589                                 spin_unlock(&info->lock);
 590                         if (!punch_hole)
 591                                 BUG_ON(page_private(subdir) > offset);
 592                 }
 593                 if (offset)
 594                         offset = 0;
 595                 else if (subdir && !page_private(subdir)) {
 596                         dir[diroff] = NULL;
 597                         nr_pages_to_free++;
 598                         list_add(&subdir->lru, &pages_to_free);
 599                 }
 600         }
 601 done1:
 602         shmem_dir_unmap(dir);
 603 done2:
 604         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 605                 /*
 606                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 607                  * may have swizzled a page in from swap since vmtruncate or
 608                  * generic_delete_inode did it, before we lowered next_index.
 609                  * Also, though shmem_getpage checks i_size before adding to
 610                  * cache, no recheck after: so fix the narrow window there too.
 611                  */
 612                 truncate_inode_pages_range(inode->i_mapping, start, end);
 613         }
 614
 615         spin_lock(&info->lock);
 616         info->flags &= ~SHMEM_TRUNCATE;
 617         info->swapped -= nr_swaps_freed;
 618         if (nr_pages_to_free)
 619                 shmem_free_blocks(inode, nr_pages_to_free);
 620         shmem_recalc_inode(inode);
 621         spin_unlock(&info->lock);
 622
 623         /*
 624          * Empty swap vector directory pages to be freed?
 625          */
 626         if (!list_empty(&pages_to_free)) {
 627                 pages_to_free.prev->next = NULL;
 628                 shmem_free_pages(pages_to_free.next);
 629         }
 630 }
 631
 632 static void shmem_truncate(struct inode *inode)
 633 {
 634         shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
 635 }
 636
 637 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 638 {
 639         struct inode *inode = dentry->d_inode;
 640         struct page *page = NULL;
 641         int error;
 642
 643         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 644                 if (attr->ia_size < inode->i_size) {
 645                         /*
 646                          * If truncating down to a partial page, then
 647                          * if that page is already allocated, hold it
 648                          * in memory until the truncation is over, so
 649                          * truncate_partial_page cannnot miss it were
 650                          * it assigned to swap.
 651                          */
 652                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 653                                 (void) shmem_getpage(inode,
 654                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 655                                                 &page, SGP_READ, NULL);
 656                         }
 657                         /*
 658                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 659                          * detect if any pages might have been added to cache
 660                          * after truncate_inode_pages.  But we needn't bother
 661                          * if it's being fully truncated to zero-length: the
 662                          * nrpages check is efficient enough in that case.
 663                          */
 664                         if (attr->ia_size) {
 665                                 struct shmem_inode_info *info = SHMEM_I(inode);
 666                                 spin_lock(&info->lock);
 667                                 info->flags &= ~SHMEM_PAGEIN;
 668                                 spin_unlock(&info->lock);
 669                         }
 670                 }
 671         }
 672
 673         error = inode_change_ok(inode, attr);
 674         if (!error)
 675                 error = inode_setattr(inode, attr);
 676 #ifdef CONFIG_TMPFS_POSIX_ACL
 677         if (!error && (attr->ia_valid & ATTR_MODE))
 678                 error = generic_acl_chmod(inode, &shmem_acl_ops);
 679 #endif
 680         if (page)
 681                 page_cache_release(page);
 682         return error;
 683 }
 684
 685 static void shmem_delete_inode(struct inode *inode)
 686 {
 687         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 688         struct shmem_inode_info *info = SHMEM_I(inode);
 689
 690         if (inode->i_op->truncate == shmem_truncate) {
 691                 truncate_inode_pages(inode->i_mapping, 0);
 692                 shmem_unacct_size(info->flags, inode->i_size);
 693                 inode->i_size = 0;
 694                 shmem_truncate(inode);
 695                 if (!list_empty(&info->swaplist)) {
 696                         spin_lock(&shmem_swaplist_lock);
 697                         list_del_init(&info->swaplist);
 698                         spin_unlock(&shmem_swaplist_lock);
 699                 }
 700         }
 701         BUG_ON(inode->i_blocks);
 702         if (sbinfo->max_inodes) {
 703                 spin_lock(&sbinfo->stat_lock);
 704                 sbinfo->free_inodes++;
 705                 spin_unlock(&sbinfo->stat_lock);
 706         }
 707         clear_inode(inode);
 708 }
 709
 710 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 711 {
 712         swp_entry_t *ptr;
 713
 714         for (ptr = dir; ptr < edir; ptr++) {
 715                 if (ptr->val == entry.val)
 716                         return ptr - dir;
 717         }
 718         return -1;
 719 }
 720
 721 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 722 {
 723         struct inode *inode;
 724         unsigned long idx;
 725         unsigned long size;
 726         unsigned long limit;
 727         unsigned long stage;
 728         struct page **dir;
 729         struct page *subdir;
 730         swp_entry_t *ptr;
 731         int offset;
 732
 733         idx = 0;
 734         ptr = info->i_direct;
 735         spin_lock(&info->lock);
 736         limit = info->next_index;
 737         size = limit;
 738         if (size > SHMEM_NR_DIRECT)
 739                 size = SHMEM_NR_DIRECT;
 740         offset = shmem_find_swp(entry, ptr, ptr+size);
 741         if (offset >= 0) {
 742                 shmem_swp_balance_unmap();
 743                 goto found;
 744         }
 745         if (!info->i_indirect)
 746                 goto lost2;
 747
 748         dir = shmem_dir_map(info->i_indirect);
 749         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 750
 751         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 752                 if (unlikely(idx == stage)) {
 753                         shmem_dir_unmap(dir-1);
 754                         dir = shmem_dir_map(info->i_indirect) +
 755                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 756                         while (!*dir) {
 757                                 dir++;
 758                                 idx += ENTRIES_PER_PAGEPAGE;
 759                                 if (idx >= limit)
 760                                         goto lost1;
 761                         }
 762                         stage = idx + ENTRIES_PER_PAGEPAGE;
 763                         subdir = *dir;
 764                         shmem_dir_unmap(dir);
 765                         dir = shmem_dir_map(subdir);
 766                 }
 767                 subdir = *dir;
 768                 if (subdir && page_private(subdir)) {
 769                         ptr = shmem_swp_map(subdir);
 770                         size = limit - idx;
 771                         if (size > ENTRIES_PER_PAGE)
 772                                 size = ENTRIES_PER_PAGE;
 773                         offset = shmem_find_swp(entry, ptr, ptr+size);
 774                         if (offset >= 0) {
 775                                 shmem_dir_unmap(dir);
 776                                 goto found;
 777                         }
 778                         shmem_swp_unmap(ptr);
 779                 }
 780         }
 781 lost1:
 782         shmem_dir_unmap(dir-1);
 783 lost2:
 784         spin_unlock(&info->lock);
 785         return 0;
 786 found:
 787         idx += offset;
 788         inode = &info->vfs_inode;
 789         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
 790                 info->flags |= SHMEM_PAGEIN;
 791                 shmem_swp_set(info, ptr + offset, 0);
 792         }
 793         shmem_swp_unmap(ptr);
 794         spin_unlock(&info->lock);
 795         /*
 796          * Decrement swap count even when the entry is left behind:
 797          * try_to_unuse will skip over mms, then reincrement count.
 798          */
 799         swap_free(entry);
 800         return 1;
 801 }
 802
 803 /*
 804  * shmem_unuse() search for an eventually swapped out shmem page.
 805  */
 806 int shmem_unuse(swp_entry_t entry, struct page *page)
 807 {
 808         struct list_head *p, *next;
 809         struct shmem_inode_info *info;
 810         int found = 0;
 811
 812         spin_lock(&shmem_swaplist_lock);
 813         list_for_each_safe(p, next, &shmem_swaplist) {
 814                 info = list_entry(p, struct shmem_inode_info, swaplist);
 815                 if (!info->swapped)
 816                         list_del_init(&info->swaplist);
 817                 else if (shmem_unuse_inode(info, entry, page)) {
 818                         /* move head to start search for next from here */
 819                         list_move_tail(&shmem_swaplist, &info->swaplist);
 820                         found = 1;
 821                         break;
 822                 }
 823         }
 824         spin_unlock(&shmem_swaplist_lock);
 825         return found;
 826 }
 827
 828 /*
 829  * Move the page from the page cache to the swap cache.
 830  */
 831 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 832 {
 833         struct shmem_inode_info *info;
 834         swp_entry_t *entry, swap;
 835         struct address_space *mapping;
 836         unsigned long index;
 837         struct inode *inode;
 838
 839         BUG_ON(!PageLocked(page));
 840         BUG_ON(page_mapped(page));
 841
 842         mapping = page->mapping;
 843         index = page->index;
 844         inode = mapping->host;
 845         info = SHMEM_I(inode);
 846         if (info->flags & VM_LOCKED)
 847                 goto redirty;
 848         swap = get_swap_page();
 849         if (!swap.val)
 850                 goto redirty;
 851
 852         spin_lock(&info->lock);
 853         shmem_recalc_inode(inode);
 854         if (index >= info->next_index) {
 855                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
 856                 goto unlock;
 857         }
 858         entry = shmem_swp_entry(info, index, NULL);
 859         BUG_ON(!entry);
 860         BUG_ON(entry->val);
 861
 862         if (move_to_swap_cache(page, swap) == 0) {
 863                 shmem_swp_set(info, entry, swap.val);
 864                 shmem_swp_unmap(entry);
 865                 spin_unlock(&info->lock);
 866                 if (list_empty(&info->swaplist)) {
 867                         spin_lock(&shmem_swaplist_lock);
 868                         /* move instead of add in case we're racing */
 869                         list_move_tail(&info->swaplist, &shmem_swaplist);
 870                         spin_unlock(&shmem_swaplist_lock);
 871                 }
 872                 unlock_page(page);
 873                 return 0;
 874         }
 875
 876         shmem_swp_unmap(entry);
 877 unlock:
 878         spin_unlock(&info->lock);
 879         swap_free(swap);
 880 redirty:
 881         set_page_dirty(page);
 882         return AOP_WRITEPAGE_ACTIVATE;  /* Return with the page locked */
 883 }
 884
 885 #ifdef CONFIG_NUMA
 886 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
 887 {
 888         char *nodelist = strchr(value, ':');
 889         int err = 1;
 890
 891         if (nodelist) {
 892                 /* NUL-terminate policy string */
 893                 *nodelist++ = '\0';
 894                 if (nodelist_parse(nodelist, *policy_nodes))
 895                         goto out;
 896         }
 897         if (!strcmp(value, "default")) {
 898                 *policy = MPOL_DEFAULT;
 899                 /* Don't allow a nodelist */
 900                 if (!nodelist)
 901                         err = 0;
 902         } else if (!strcmp(value, "prefer")) {
 903                 *policy = MPOL_PREFERRED;
 904                 /* Insist on a nodelist of one node only */
 905                 if (nodelist) {
 906                         char *rest = nodelist;
 907                         while (isdigit(*rest))
 908                                 rest++;
 909                         if (!*rest)
 910                                 err = 0;
 911                 }
 912         } else if (!strcmp(value, "bind")) {
 913                 *policy = MPOL_BIND;
 914                 /* Insist on a nodelist */
 915                 if (nodelist)
 916                         err = 0;
 917         } else if (!strcmp(value, "interleave")) {
 918                 *policy = MPOL_INTERLEAVE;
 919                 /* Default to nodes online if no nodelist */
 920                 if (!nodelist)
 921                         *policy_nodes = node_online_map;
 922                 err = 0;
 923         }
 924 out:
 925         /* Restore string for error message */
 926         if (nodelist)
 927                 *--nodelist = ':';
 928         return err;
 929 }
 930
 931 static struct page *shmem_swapin_async(struct shared_policy *p,
 932                                        swp_entry_t entry, unsigned long idx)
 933 {
 934         struct page *page;
 935         struct vm_area_struct pvma;
 936
 937         /* Create a pseudo vma that just contains the policy */
 938         memset(&pvma, 0, sizeof(struct vm_area_struct));
 939         pvma.vm_end = PAGE_SIZE;
 940         pvma.vm_pgoff = idx;
 941         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
 942         page = read_swap_cache_async(entry, &pvma, 0);
 943         mpol_free(pvma.vm_policy);
 944         return page;
 945 }
 946
 947 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 948                           unsigned long idx)
 949 {
 950         struct shared_policy *p = &info->policy;
 951         int i, num;
 952         struct page *page;
 953         unsigned long offset;
 954
 955         num = valid_swaphandles(entry, &offset);
 956         for (i = 0; i < num; offset++, i++) {
 957                 page = shmem_swapin_async(p,
 958                                 swp_entry(swp_type(entry), offset), idx);
 959                 if (!page)
 960                         break;
 961                 page_cache_release(page);
 962         }
 963         lru_add_drain();        /* Push any new pages onto the LRU now */
 964         return shmem_swapin_async(p, entry, idx);
 965 }
 966
 967 static struct page *
 968 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
 969                  unsigned long idx)
 970 {
 971         struct vm_area_struct pvma;
 972         struct page *page;
 973
 974         memset(&pvma, 0, sizeof(struct vm_area_struct));
 975         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
 976         pvma.vm_pgoff = idx;
 977         pvma.vm_end = PAGE_SIZE;
 978         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
 979         mpol_free(pvma.vm_policy);
 980         return page;
 981 }
 982 #else
 983 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
 984 {
 985         return 1;
 986 }
 987
 988 static inline struct page *
 989 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 990 {
 991         swapin_readahead(entry, 0, NULL);
 992         return read_swap_cache_async(entry, NULL, 0);
 993 }
 994
 995 static inline struct page *
 996 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
 997 {
 998         return alloc_page(gfp | __GFP_ZERO);
 999 }
1000 #endif
1001
1002 /*
1003  * shmem_getpage - either get the page from swap or allocate a new one
1004  *
1005  * If we allocate a new one we do not mark it dirty. That's up to the
1006  * vm. If we swap it in we mark it dirty since we also free the swap
1007  * entry since a page cannot live in both the swap and page cache
1008  */
1009 static int shmem_getpage(struct inode *inode, unsigned long idx,
1010                         struct page **pagep, enum sgp_type sgp, int *type)
1011 {
1012         struct address_space *mapping = inode->i_mapping;
1013         struct shmem_inode_info *info = SHMEM_I(inode);
1014         struct shmem_sb_info *sbinfo;
1015         struct page *filepage = *pagep;
1016         struct page *swappage;
1017         swp_entry_t *entry;
1018         swp_entry_t swap;
1019         int error;
1020
1021         if (idx >= SHMEM_MAX_INDEX)
1022                 return -EFBIG;
1023         /*
1024          * Normally, filepage is NULL on entry, and either found
1025          * uptodate immediately, or allocated and zeroed, or read
1026          * in under swappage, which is then assigned to filepage.
1027          * But shmem_prepare_write passes in a locked filepage,
1028          * which may be found not uptodate by other callers too,
1029          * and may need to be copied from the swappage read in.
1030          */
1031 repeat:
1032         if (!filepage)
1033                 filepage = find_lock_page(mapping, idx);
1034         if (filepage && PageUptodate(filepage))
1035                 goto done;
1036         error = 0;
1037         if (sgp == SGP_QUICK)
1038                 goto failed;
1039
1040         spin_lock(&info->lock);
1041         shmem_recalc_inode(inode);
1042         entry = shmem_swp_alloc(info, idx, sgp);
1043         if (IS_ERR(entry)) {
1044                 spin_unlock(&info->lock);
1045                 error = PTR_ERR(entry);
1046                 goto failed;
1047         }
1048         swap = *entry;
1049
1050         if (swap.val) {
1051                 /* Look it up and read it in.. */
1052                 swappage = lookup_swap_cache(swap);
1053                 if (!swappage) {
1054                         shmem_swp_unmap(entry);
1055                         /* here we actually do the io */
1056                         if (type && *type == VM_FAULT_MINOR) {
1057                                 __count_vm_event(PGMAJFAULT);
1058                                 *type = VM_FAULT_MAJOR;
1059                         }
1060                         spin_unlock(&info->lock);
1061                         swappage = shmem_swapin(info, swap, idx);
1062                         if (!swappage) {
1063                                 spin_lock(&info->lock);
1064                                 entry = shmem_swp_alloc(info, idx, sgp);
1065                                 if (IS_ERR(entry))
1066                                         error = PTR_ERR(entry);
1067                                 else {
1068                                         if (entry->val == swap.val)
1069                                                 error = -ENOMEM;
1070                                         shmem_swp_unmap(entry);
1071                                 }
1072                                 spin_unlock(&info->lock);
1073                                 if (error)
1074                                         goto failed;
1075                                 goto repeat;
1076                         }
1077                         wait_on_page_locked(swappage);
1078                         page_cache_release(swappage);
1079                         goto repeat;
1080                 }
1081
1082                 /* We have to do this with page locked to prevent races */
1083                 if (TestSetPageLocked(swappage)) {
1084                         shmem_swp_unmap(entry);
1085                         spin_unlock(&info->lock);
1086                         wait_on_page_locked(swappage);
1087                         page_cache_release(swappage);
1088                         goto repeat;
1089                 }
1090                 if (PageWriteback(swappage)) {
1091                         shmem_swp_unmap(entry);
1092                         spin_unlock(&info->lock);
1093                         wait_on_page_writeback(swappage);
1094                         unlock_page(swappage);
1095                         page_cache_release(swappage);
1096                         goto repeat;
1097                 }
1098                 if (!PageUptodate(swappage)) {
1099                         shmem_swp_unmap(entry);
1100                         spin_unlock(&info->lock);
1101                         unlock_page(swappage);
1102                         page_cache_release(swappage);
1103                         error = -EIO;
1104                         goto failed;
1105                 }
1106
1107                 if (filepage) {
1108                         shmem_swp_set(info, entry, 0);
1109                         shmem_swp_unmap(entry);
1110                         delete_from_swap_cache(swappage);
1111                         spin_unlock(&info->lock);
1112                         copy_highpage(filepage, swappage);
1113                         unlock_page(swappage);
1114                         page_cache_release(swappage);
1115                         flush_dcache_page(filepage);
1116                         SetPageUptodate(filepage);
1117                         set_page_dirty(filepage);
1118                         swap_free(swap);
1119                 } else if (!(error = move_from_swap_cache(
1120                                 swappage, idx, mapping))) {
1121                         info->flags |= SHMEM_PAGEIN;
1122                         shmem_swp_set(info, entry, 0);
1123                         shmem_swp_unmap(entry);
1124                         spin_unlock(&info->lock);
1125                         filepage = swappage;
1126                         swap_free(swap);
1127                 } else {
1128                         shmem_swp_unmap(entry);
1129                         spin_unlock(&info->lock);
1130                         unlock_page(swappage);
1131                         page_cache_release(swappage);
1132                         if (error == -ENOMEM) {
1133                                 /* let kswapd refresh zone for GFP_ATOMICs */
1134                                 blk_congestion_wait(WRITE, HZ/50);
1135                         }
1136                         goto repeat;
1137                 }
1138         } else if (sgp == SGP_READ && !filepage) {
1139                 shmem_swp_unmap(entry);
1140                 filepage = find_get_page(mapping, idx);
1141                 if (filepage &&
1142                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1143                         spin_unlock(&info->lock);
1144                         wait_on_page_locked(filepage);
1145                         page_cache_release(filepage);
1146                         filepage = NULL;
1147                         goto repeat;
1148                 }
1149                 spin_unlock(&info->lock);
1150         } else {
1151                 shmem_swp_unmap(entry);
1152                 sbinfo = SHMEM_SB(inode->i_sb);
1153                 if (sbinfo->max_blocks) {
1154                         spin_lock(&sbinfo->stat_lock);
1155                         if (sbinfo->free_blocks == 0 ||
1156                             shmem_acct_block(info->flags)) {
1157                                 spin_unlock(&sbinfo->stat_lock);
1158                                 spin_unlock(&info->lock);
1159                                 error = -ENOSPC;
1160                                 goto failed;
1161                         }
1162                         sbinfo->free_blocks--;
1163                         inode->i_blocks += BLOCKS_PER_PAGE;
1164                         spin_unlock(&sbinfo->stat_lock);
1165                 } else if (shmem_acct_block(info->flags)) {
1166                         spin_unlock(&info->lock);
1167                         error = -ENOSPC;
1168                         goto failed;
1169                 }
1170
1171                 if (!filepage) {
1172                         spin_unlock(&info->lock);
1173                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1174                                                     info,
1175                                                     idx);
1176                         if (!filepage) {
1177                                 shmem_unacct_blocks(info->flags, 1);
1178                                 shmem_free_blocks(inode, 1);
1179                                 error = -ENOMEM;
1180                                 goto failed;
1181                         }
1182
1183                         spin_lock(&info->lock);
1184                         entry = shmem_swp_alloc(info, idx, sgp);
1185                         if (IS_ERR(entry))
1186                                 error = PTR_ERR(entry);
1187                         else {
1188                                 swap = *entry;
1189                                 shmem_swp_unmap(entry);
1190                         }
1191                         if (error || swap.val || 0 != add_to_page_cache_lru(
1192                                         filepage, mapping, idx, GFP_ATOMIC)) {
1193                                 spin_unlock(&info->lock);
1194                                 page_cache_release(filepage);
1195                                 shmem_unacct_blocks(info->flags, 1);
1196                                 shmem_free_blocks(inode, 1);
1197                                 filepage = NULL;
1198                                 if (error)
1199                                         goto failed;
1200                                 goto repeat;
1201                         }
1202                         info->flags |= SHMEM_PAGEIN;
1203                 }
1204
1205                 info->alloced++;
1206                 spin_unlock(&info->lock);
1207                 flush_dcache_page(filepage);
1208                 SetPageUptodate(filepage);
1209         }
1210 done:
1211         if (*pagep != filepage) {
1212                 unlock_page(filepage);
1213                 *pagep = filepage;
1214         }
1215         return 0;
1216
1217 failed:
1218         if (*pagep != filepage) {
1219                 unlock_page(filepage);
1220                 page_cache_release(filepage);
1221         }
1222         return error;
1223 }
1224
1225 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1226 {
1227         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1228         struct page *page = NULL;
1229         unsigned long idx;
1230         int error;
1231
1232         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1233         idx += vma->vm_pgoff;
1234         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1235         if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1236                 return NOPAGE_SIGBUS;
1237
1238         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1239         if (error)
1240                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1241
1242         mark_page_accessed(page);
1243         return page;
1244 }
1245
1246 static int shmem_populate(struct vm_area_struct *vma,
1247         unsigned long addr, unsigned long len,
1248         pgprot_t prot, unsigned long pgoff, int nonblock)
1249 {
1250         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1251         struct mm_struct *mm = vma->vm_mm;
1252         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1253         unsigned long size;
1254
1255         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1256         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1257                 return -EINVAL;
1258
1259         while ((long) len > 0) {
1260                 struct page *page = NULL;
1261                 int err;
1262                 /*
1263                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1264                  */
1265                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1266                 if (err)
1267                         return err;
1268                 /* Page may still be null, but only if nonblock was set. */
1269                 if (page) {
1270                         mark_page_accessed(page);
1271                         err = install_page(mm, vma, addr, page, prot);
1272                         if (err) {
1273                                 page_cache_release(page);
1274                                 return err;
1275                         }
1276                 } else if (vma->vm_flags & VM_NONLINEAR) {
1277                         /* No page was found just because we can't read it in
1278                          * now (being here implies nonblock != 0), but the page
1279                          * may exist, so set the PTE to fault it in later. */
1280                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1281                         if (err)
1282                                 return err;
1283                 }
1284
1285                 len -= PAGE_SIZE;
1286                 addr += PAGE_SIZE;
1287                 pgoff++;
1288         }
1289         return 0;
1290 }
1291
1292 #ifdef CONFIG_NUMA
1293 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1294 {
1295         struct inode *i = vma->vm_file->f_dentry->d_inode;
1296         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1297 }
1298
1299 struct mempolicy *
1300 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1301 {
1302         struct inode *i = vma->vm_file->f_dentry->d_inode;
1303         unsigned long idx;
1304
1305         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1306         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1307 }
1308 #endif
1309
1310 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1311 {
1312         struct inode *inode = file->f_dentry->d_inode;
1313         struct shmem_inode_info *info = SHMEM_I(inode);
1314         int retval = -ENOMEM;
1315
1316         spin_lock(&info->lock);
1317         if (lock && !(info->flags & VM_LOCKED)) {
1318                 if (!user_shm_lock(inode->i_size, user))
1319                         goto out_nomem;
1320                 info->flags |= VM_LOCKED;
1321         }
1322         if (!lock && (info->flags & VM_LOCKED) && user) {
1323                 user_shm_unlock(inode->i_size, user);
1324                 info->flags &= ~VM_LOCKED;
1325         }
1326         retval = 0;
1327 out_nomem:
1328         spin_unlock(&info->lock);
1329         return retval;
1330 }
1331
1332 int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1333 {
1334         file_accessed(file);
1335         vma->vm_ops = &shmem_vm_ops;
1336         return 0;
1337 }
1338
1339 static struct inode *
1340 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1341 {
1342         struct inode *inode;
1343         struct shmem_inode_info *info;
1344         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1345
1346         if (sbinfo->max_inodes) {
1347                 spin_lock(&sbinfo->stat_lock);
1348                 if (!sbinfo->free_inodes) {
1349                         spin_unlock(&sbinfo->stat_lock);
1350                         return NULL;
1351                 }
1352                 sbinfo->free_inodes--;
1353                 spin_unlock(&sbinfo->stat_lock);
1354         }
1355
1356         inode = new_inode(sb);
1357         if (inode) {
1358                 inode->i_mode = mode;
1359                 inode->i_uid = current->fsuid;
1360                 inode->i_gid = current->fsgid;
1361                 inode->i_blocks = 0;
1362                 inode->i_mapping->a_ops = &shmem_aops;
1363                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1364                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1365                 inode->i_generation = get_seconds();
1366                 info = SHMEM_I(inode);
1367                 memset(info, 0, (char *)inode - (char *)info);
1368                 spin_lock_init(&info->lock);
1369                 INIT_LIST_HEAD(&info->swaplist);
1370
1371                 switch (mode & S_IFMT) {
1372                 default:
1373                         inode->i_op = &shmem_special_inode_operations;
1374                         init_special_inode(inode, mode, dev);
1375                         break;
1376                 case S_IFREG:
1377                         inode->i_op = &shmem_inode_operations;
1378                         inode->i_fop = &shmem_file_operations;
1379                         mpol_shared_policy_init(&info->policy, sbinfo->policy,
1380                                                         &sbinfo->policy_nodes);
1381                         break;
1382                 case S_IFDIR:
1383                         inc_nlink(inode);
1384                         /* Some things misbehave if size == 0 on a directory */
1385                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1386                         inode->i_op = &shmem_dir_inode_operations;
1387                         inode->i_fop = &simple_dir_operations;
1388                         break;
1389                 case S_IFLNK:
1390                         /*
1391                          * Must not load anything in the rbtree,
1392                          * mpol_free_shared_policy will not be called.
1393                          */
1394                         mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1395                                                 NULL);
1396                         break;
1397                 }
1398         } else if (sbinfo->max_inodes) {
1399                 spin_lock(&sbinfo->stat_lock);
1400                 sbinfo->free_inodes++;
1401                 spin_unlock(&sbinfo->stat_lock);
1402         }
1403         return inode;
1404 }
1405
1406 #ifdef CONFIG_TMPFS
1407 static struct inode_operations shmem_symlink_inode_operations;
1408 static struct inode_operations shmem_symlink_inline_operations;
1409
1410 /*
1411  * Normally tmpfs makes no use of shmem_prepare_write, but it
1412  * lets a tmpfs file be used read-write below the loop driver.
1413  */
1414 static int
1415 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1416 {
1417         struct inode *inode = page->mapping->host;
1418         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1419 }
1420
1421 static ssize_t
1422 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1423 {
1424         struct inode    *inode = file->f_dentry->d_inode;
1425         loff_t          pos;
1426         unsigned long   written;
1427         ssize_t         err;
1428
1429         if ((ssize_t) count < 0)
1430                 return -EINVAL;
1431
1432         if (!access_ok(VERIFY_READ, buf, count))
1433                 return -EFAULT;
1434
1435         mutex_lock(&inode->i_mutex);
1436
1437         pos = *ppos;
1438         written = 0;
1439
1440         err = generic_write_checks(file, &pos, &count, 0);
1441         if (err || !count)
1442                 goto out;
1443
1444         err = remove_suid(file->f_dentry);
1445         if (err)
1446                 goto out;
1447
1448         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1449
1450         do {
1451                 struct page *page = NULL;
1452                 unsigned long bytes, index, offset;
1453                 char *kaddr;
1454                 int left;
1455
1456                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1457                 index = pos >> PAGE_CACHE_SHIFT;
1458                 bytes = PAGE_CACHE_SIZE - offset;
1459                 if (bytes > count)
1460                         bytes = count;
1461
1462                 /*
1463                  * We don't hold page lock across copy from user -
1464                  * what would it guard against? - so no deadlock here.
1465                  * But it still may be a good idea to prefault below.
1466                  */
1467
1468                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1469                 if (err)
1470                         break;
1471
1472                 left = bytes;
1473                 if (PageHighMem(page)) {
1474                         volatile unsigned char dummy;
1475                         __get_user(dummy, buf);
1476                         __get_user(dummy, buf + bytes - 1);
1477
1478                         kaddr = kmap_atomic(page, KM_USER0);
1479                         left = __copy_from_user_inatomic(kaddr + offset,
1480                                                         buf, bytes);
1481                         kunmap_atomic(kaddr, KM_USER0);
1482                 }
1483                 if (left) {
1484                         kaddr = kmap(page);
1485                         left = __copy_from_user(kaddr + offset, buf, bytes);
1486                         kunmap(page);
1487                 }
1488
1489                 written += bytes;
1490                 count -= bytes;
1491                 pos += bytes;
1492                 buf += bytes;
1493                 if (pos > inode->i_size)
1494                         i_size_write(inode, pos);
1495
1496                 flush_dcache_page(page);
1497                 set_page_dirty(page);
1498                 mark_page_accessed(page);
1499                 page_cache_release(page);
1500
1501                 if (left) {
1502                         pos -= left;
1503                         written -= left;
1504                         err = -EFAULT;
1505                         break;
1506                 }
1507
1508                 /*
1509                  * Our dirty pages are not counted in nr_dirty,
1510                  * and we do not attempt to balance dirty pages.
1511                  */
1512
1513                 cond_resched();
1514         } while (count);
1515
1516         *ppos = pos;
1517         if (written)
1518                 err = written;
1519 out:
1520         mutex_unlock(&inode->i_mutex);
1521         return err;
1522 }
1523
1524 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1525 {
1526         struct inode *inode = filp->f_dentry->d_inode;
1527         struct address_space *mapping = inode->i_mapping;
1528         unsigned long index, offset;
1529
1530         index = *ppos >> PAGE_CACHE_SHIFT;
1531         offset = *ppos & ~PAGE_CACHE_MASK;
1532
1533         for (;;) {
1534                 struct page *page = NULL;
1535                 unsigned long end_index, nr, ret;
1536                 loff_t i_size = i_size_read(inode);
1537
1538                 end_index = i_size >> PAGE_CACHE_SHIFT;
1539                 if (index > end_index)
1540                         break;
1541                 if (index == end_index) {
1542                         nr = i_size & ~PAGE_CACHE_MASK;
1543                         if (nr <= offset)
1544                                 break;
1545                 }
1546
1547                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1548                 if (desc->error) {
1549                         if (desc->error == -EINVAL)
1550                                 desc->error = 0;
1551                         break;
1552                 }
1553
1554                 /*
1555                  * We must evaluate after, since reads (unlike writes)
1556                  * are called without i_mutex protection against truncate
1557                  */
1558                 nr = PAGE_CACHE_SIZE;
1559                 i_size = i_size_read(inode);
1560                 end_index = i_size >> PAGE_CACHE_SHIFT;
1561                 if (index == end_index) {
1562                         nr = i_size & ~PAGE_CACHE_MASK;
1563                         if (nr <= offset) {
1564                                 if (page)
1565                                         page_cache_release(page);
1566                                 break;
1567                         }
1568                 }
1569                 nr -= offset;
1570
1571                 if (page) {
1572                         /*
1573                          * If users can be writing to this page using arbitrary
1574                          * virtual addresses, take care about potential aliasing
1575                          * before reading the page on the kernel side.
1576                          */
1577                         if (mapping_writably_mapped(mapping))
1578                                 flush_dcache_page(page);
1579                         /*
1580                          * Mark the page accessed if we read the beginning.
1581                          */
1582                         if (!offset)
1583                                 mark_page_accessed(page);
1584                 } else {
1585                         page = ZERO_PAGE(0);
1586                         page_cache_get(page);
1587                 }
1588
1589                 /*
1590                  * Ok, we have the page, and it's up-to-date, so
1591                  * now we can copy it to user space...
1592                  *
1593                  * The actor routine returns how many bytes were actually used..
1594                  * NOTE! This may not be the same as how much of a user buffer
1595                  * we filled up (we may be padding etc), so we can only update
1596                  * "pos" here (the actor routine has to update the user buffer
1597                  * pointers and the remaining count).
1598                  */
1599                 ret = actor(desc, page, offset, nr);
1600                 offset += ret;
1601                 index += offset >> PAGE_CACHE_SHIFT;
1602                 offset &= ~PAGE_CACHE_MASK;
1603
1604                 page_cache_release(page);
1605                 if (ret != nr || !desc->count)
1606                         break;
1607
1608                 cond_resched();
1609         }
1610
1611         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1612         file_accessed(filp);
1613 }
1614
1615 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1616 {
1617         read_descriptor_t desc;
1618
1619         if ((ssize_t) count < 0)
1620                 return -EINVAL;
1621         if (!access_ok(VERIFY_WRITE, buf, count))
1622                 return -EFAULT;
1623         if (!count)
1624                 return 0;
1625
1626         desc.written = 0;
1627         desc.count = count;
1628         desc.arg.buf = buf;
1629         desc.error = 0;
1630
1631         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1632         if (desc.written)
1633                 return desc.written;
1634         return desc.error;
1635 }
1636
1637 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1638                          size_t count, read_actor_t actor, void *target)
1639 {
1640         read_descriptor_t desc;
1641
1642         if (!count)
1643                 return 0;
1644
1645         desc.written = 0;
1646         desc.count = count;
1647         desc.arg.data = target;
1648         desc.error = 0;
1649
1650         do_shmem_file_read(in_file, ppos, &desc, actor);
1651         if (desc.written)
1652                 return desc.written;
1653         return desc.error;
1654 }
1655
1656 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1657 {
1658         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1659
1660         buf->f_type = TMPFS_MAGIC;
1661         buf->f_bsize = PAGE_CACHE_SIZE;
1662         buf->f_namelen = NAME_MAX;
1663         spin_lock(&sbinfo->stat_lock);
1664         if (sbinfo->max_blocks) {
1665                 buf->f_blocks = sbinfo->max_blocks;
1666                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1667         }
1668         if (sbinfo->max_inodes) {
1669                 buf->f_files = sbinfo->max_inodes;
1670                 buf->f_ffree = sbinfo->free_inodes;
1671         }
1672         /* else leave those fields 0 like simple_statfs */
1673         spin_unlock(&sbinfo->stat_lock);
1674         return 0;
1675 }
1676
1677 /*
1678  * File creation. Allocate an inode, and we're done..
1679  */
1680 static int
1681 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1682 {
1683         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1684         int error = -ENOSPC;
1685
1686         if (inode) {
1687                 error = security_inode_init_security(inode, dir, NULL, NULL,
1688                                                      NULL);
1689                 if (error) {
1690                         if (error != -EOPNOTSUPP) {
1691                                 iput(inode);
1692                                 return error;
1693                         }
1694                 }
1695                 error = shmem_acl_init(inode, dir);
1696                 if (error) {
1697                         iput(inode);
1698                         return error;
1699                 }
1700                 if (dir->i_mode & S_ISGID) {
1701                         inode->i_gid = dir->i_gid;
1702                         if (S_ISDIR(mode))
1703                                 inode->i_mode |= S_ISGID;
1704                 }
1705                 dir->i_size += BOGO_DIRENT_SIZE;
1706                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1707                 d_instantiate(dentry, inode);
1708                 dget(dentry); /* Extra count - pin the dentry in core */
1709         }
1710         return error;
1711 }
1712
1713 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1714 {
1715         int error;
1716
1717         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1718                 return error;
1719         inc_nlink(dir);
1720         return 0;
1721 }
1722
1723 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1724                 struct nameidata *nd)
1725 {
1726         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1727 }
1728
1729 /*
1730  * Link a file..
1731  */
1732 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1733 {
1734         struct inode *inode = old_dentry->d_inode;
1735         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1736
1737         /*
1738          * No ordinary (disk based) filesystem counts links as inodes;
1739          * but each new link needs a new dentry, pinning lowmem, and
1740          * tmpfs dentries cannot be pruned until they are unlinked.
1741          */
1742         if (sbinfo->max_inodes) {
1743                 spin_lock(&sbinfo->stat_lock);
1744                 if (!sbinfo->free_inodes) {
1745                         spin_unlock(&sbinfo->stat_lock);
1746                         return -ENOSPC;
1747                 }
1748                 sbinfo->free_inodes--;
1749                 spin_unlock(&sbinfo->stat_lock);
1750         }
1751
1752         dir->i_size += BOGO_DIRENT_SIZE;
1753         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1754         inc_nlink(inode);
1755         atomic_inc(&inode->i_count);    /* New dentry reference */
1756         dget(dentry);           /* Extra pinning count for the created dentry */
1757         d_instantiate(dentry, inode);
1758         return 0;
1759 }
1760
1761 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1762 {
1763         struct inode *inode = dentry->d_inode;
1764
1765         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1766                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1767                 if (sbinfo->max_inodes) {
1768                         spin_lock(&sbinfo->stat_lock);
1769                         sbinfo->free_inodes++;
1770                         spin_unlock(&sbinfo->stat_lock);
1771                 }
1772         }
1773
1774         dir->i_size -= BOGO_DIRENT_SIZE;
1775         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1776         drop_nlink(inode);
1777         dput(dentry);   /* Undo the count from "create" - this does all the work */
1778         return 0;
1779 }
1780
1781 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1782 {
1783         if (!simple_empty(dentry))
1784                 return -ENOTEMPTY;
1785
1786         drop_nlink(dentry->d_inode);
1787         drop_nlink(dir);
1788         return shmem_unlink(dir, dentry);
1789 }
1790
1791 /*
1792  * The VFS layer already does all the dentry stuff for rename,
1793  * we just have to decrement the usage count for the target if
1794  * it exists so that the VFS layer correctly free's it when it
1795  * gets overwritten.
1796  */
1797 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1798 {
1799         struct inode *inode = old_dentry->d_inode;
1800         int they_are_dirs = S_ISDIR(inode->i_mode);
1801
1802         if (!simple_empty(new_dentry))
1803                 return -ENOTEMPTY;
1804
1805         if (new_dentry->d_inode) {
1806                 (void) shmem_unlink(new_dir, new_dentry);
1807                 if (they_are_dirs)
1808                         drop_nlink(old_dir);
1809         } else if (they_are_dirs) {
1810                 drop_nlink(old_dir);
1811                 inc_nlink(new_dir);
1812         }
1813
1814         old_dir->i_size -= BOGO_DIRENT_SIZE;
1815         new_dir->i_size += BOGO_DIRENT_SIZE;
1816         old_dir->i_ctime = old_dir->i_mtime =
1817         new_dir->i_ctime = new_dir->i_mtime =
1818         inode->i_ctime = CURRENT_TIME;
1819         return 0;
1820 }
1821
1822 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1823 {
1824         int error;
1825         int len;
1826         struct inode *inode;
1827         struct page *page = NULL;
1828         char *kaddr;
1829         struct shmem_inode_info *info;
1830
1831         len = strlen(symname) + 1;
1832         if (len > PAGE_CACHE_SIZE)
1833                 return -ENAMETOOLONG;
1834
1835         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1836         if (!inode)
1837                 return -ENOSPC;
1838
1839         error = security_inode_init_security(inode, dir, NULL, NULL,
1840                                              NULL);
1841         if (error) {
1842                 if (error != -EOPNOTSUPP) {
1843                         iput(inode);
1844                         return error;
1845                 }
1846                 error = 0;
1847         }
1848
1849         info = SHMEM_I(inode);
1850         inode->i_size = len-1;
1851         if (len <= (char *)inode - (char *)info) {
1852                 /* do it inline */
1853                 memcpy(info, symname, len);
1854                 inode->i_op = &shmem_symlink_inline_operations;
1855         } else {
1856                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1857                 if (error) {
1858                         iput(inode);
1859                         return error;
1860                 }
1861                 inode->i_op = &shmem_symlink_inode_operations;
1862                 kaddr = kmap_atomic(page, KM_USER0);
1863                 memcpy(kaddr, symname, len);
1864                 kunmap_atomic(kaddr, KM_USER0);
1865                 set_page_dirty(page);
1866                 page_cache_release(page);
1867         }
1868         if (dir->i_mode & S_ISGID)
1869                 inode->i_gid = dir->i_gid;
1870         dir->i_size += BOGO_DIRENT_SIZE;
1871         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1872         d_instantiate(dentry, inode);
1873         dget(dentry);
1874         return 0;
1875 }
1876
1877 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1878 {
1879         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1880         return NULL;
1881 }
1882
1883 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1884 {
1885         struct page *page = NULL;
1886         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1887         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1888         return page;
1889 }
1890
1891 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1892 {
1893         if (!IS_ERR(nd_get_link(nd))) {
1894                 struct page *page = cookie;
1895                 kunmap(page);
1896                 mark_page_accessed(page);
1897                 page_cache_release(page);
1898         }
1899 }
1900
1901 static struct inode_operations shmem_symlink_inline_operations = {
1902         .readlink       = generic_readlink,
1903         .follow_link    = shmem_follow_link_inline,
1904 };
1905
1906 static struct inode_operations shmem_symlink_inode_operations = {
1907         .truncate       = shmem_truncate,
1908         .readlink       = generic_readlink,
1909         .follow_link    = shmem_follow_link,
1910         .put_link       = shmem_put_link,
1911 };
1912
1913 #ifdef CONFIG_TMPFS_POSIX_ACL
1914 /**
1915  * Superblocks without xattr inode operations will get security.* xattr
1916  * support from the VFS "for free". As soon as we have any other xattrs
1917  * like ACLs, we also need to implement the security.* handlers at
1918  * filesystem level, though.
1919  */
1920
1921 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1922                                         size_t list_len, const char *name,
1923                                         size_t name_len)
1924 {
1925         return security_inode_listsecurity(inode, list, list_len);
1926 }
1927
1928 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1929                                     void *buffer, size_t size)
1930 {
1931         if (strcmp(name, "") == 0)
1932                 return -EINVAL;
1933         return security_inode_getsecurity(inode, name, buffer, size,
1934                                           -EOPNOTSUPP);
1935 }
1936
1937 static int shmem_xattr_security_set(struct inode *inode, const char *name,
1938                                     const void *value, size_t size, int flags)
1939 {
1940         if (strcmp(name, "") == 0)
1941                 return -EINVAL;
1942         return security_inode_setsecurity(inode, name, value, size, flags);
1943 }
1944
1945 struct xattr_handler shmem_xattr_security_handler = {
1946         .prefix = XATTR_SECURITY_PREFIX,
1947         .list   = shmem_xattr_security_list,
1948         .get    = shmem_xattr_security_get,
1949         .set    = shmem_xattr_security_set,
1950 };
1951
1952 static struct xattr_handler *shmem_xattr_handlers[] = {
1953         &shmem_xattr_acl_access_handler,
1954         &shmem_xattr_acl_default_handler,
1955         &shmem_xattr_security_handler,
1956         NULL
1957 };
1958 #endif
1959
1960 static struct dentry *shmem_get_parent(struct dentry *child)
1961 {
1962         return ERR_PTR(-ESTALE);
1963 }
1964
1965 static int shmem_match(struct inode *ino, void *vfh)
1966 {
1967         __u32 *fh = vfh;
1968         __u64 inum = fh[2];
1969         inum = (inum << 32) | fh[1];
1970         return ino->i_ino == inum && fh[0] == ino->i_generation;
1971 }
1972
1973 static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
1974 {
1975         struct dentry *de = NULL;
1976         struct inode *inode;
1977         __u32 *fh = vfh;
1978         __u64 inum = fh[2];
1979         inum = (inum << 32) | fh[1];
1980
1981         inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
1982         if (inode) {
1983                 de = d_find_alias(inode);
1984                 iput(inode);
1985         }
1986
1987         return de? de: ERR_PTR(-ESTALE);
1988 }
1989
1990 static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
1991                 int len, int type,
1992                 int (*acceptable)(void *context, struct dentry *de),
1993                 void *context)
1994 {
1995         if (len < 3)
1996                 return ERR_PTR(-ESTALE);
1997
1998         return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
1999                                                         context);
2000 }
2001
2002 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2003                                 int connectable)
2004 {
2005         struct inode *inode = dentry->d_inode;
2006
2007         if (*len < 3)
2008                 return 255;
2009
2010         if (hlist_unhashed(&inode->i_hash)) {
2011                 /* Unfortunately insert_inode_hash is not idempotent,
2012                  * so as we hash inodes here rather than at creation
2013                  * time, we need a lock to ensure we only try
2014                  * to do it once
2015                  */
2016                 static DEFINE_SPINLOCK(lock);
2017                 spin_lock(&lock);
2018                 if (hlist_unhashed(&inode->i_hash))
2019                         __insert_inode_hash(inode,
2020                                             inode->i_ino + inode->i_generation);
2021                 spin_unlock(&lock);
2022         }
2023
2024         fh[0] = inode->i_generation;
2025         fh[1] = inode->i_ino;
2026         fh[2] = ((__u64)inode->i_ino) >> 32;
2027
2028         *len = 3;
2029         return 1;
2030 }
2031
2032 static struct export_operations shmem_export_ops = {
2033         .get_parent     = shmem_get_parent,
2034         .get_dentry     = shmem_get_dentry,
2035         .encode_fh      = shmem_encode_fh,
2036         .decode_fh      = shmem_decode_fh,
2037 };
2038
2039 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2040         gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2041         int *policy, nodemask_t *policy_nodes)
2042 {
2043         char *this_char, *value, *rest;
2044
2045         while (options != NULL) {
2046                 this_char = options;
2047                 for (;;) {
2048                         /*
2049                          * NUL-terminate this option: unfortunately,
2050                          * mount options form a comma-separated list,
2051                          * but mpol's nodelist may also contain commas.
2052                          */
2053                         options = strchr(options, ',');
2054                         if (options == NULL)
2055                                 break;
2056                         options++;
2057                         if (!isdigit(*options)) {
2058                                 options[-1] = '\0';
2059                                 break;
2060                         }
2061                 }
2062                 if (!*this_char)
2063                         continue;
2064                 if ((value = strchr(this_char,'=')) != NULL) {
2065                         *value++ = 0;
2066                 } else {
2067                         printk(KERN_ERR
2068                             "tmpfs: No value for mount option '%s'\n",
2069                             this_char);
2070                         return 1;
2071                 }
2072
2073                 if (!strcmp(this_char,"size")) {
2074                         unsigned long long size;
2075                         size = memparse(value,&rest);
2076                         if (*rest == '%') {
2077                                 size <<= PAGE_SHIFT;
2078                                 size *= totalram_pages;
2079                                 do_div(size, 100);
2080                                 rest++;
2081                         }
2082                         if (*rest)
2083                                 goto bad_val;
2084                         *blocks = size >> PAGE_CACHE_SHIFT;
2085                 } else if (!strcmp(this_char,"nr_blocks")) {
2086                         *blocks = memparse(value,&rest);
2087                         if (*rest)
2088                                 goto bad_val;
2089                 } else if (!strcmp(this_char,"nr_inodes")) {
2090                         *inodes = memparse(value,&rest);
2091                         if (*rest)
2092                                 goto bad_val;
2093                 } else if (!strcmp(this_char,"mode")) {
2094                         if (!mode)
2095                                 continue;
2096                         *mode = simple_strtoul(value,&rest,8);
2097                         if (*rest)
2098                                 goto bad_val;
2099                 } else if (!strcmp(this_char,"uid")) {
2100                         if (!uid)
2101                                 continue;
2102                         *uid = simple_strtoul(value,&rest,0);
2103                         if (*rest)
2104                                 goto bad_val;
2105                 } else if (!strcmp(this_char,"gid")) {
2106                         if (!gid)
2107                                 continue;
2108                         *gid = simple_strtoul(value,&rest,0);
2109                         if (*rest)
2110                                 goto bad_val;
2111                 } else if (!strcmp(this_char,"mpol")) {
2112                         if (shmem_parse_mpol(value,policy,policy_nodes))
2113                                 goto bad_val;
2114                 } else {
2115                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2116                                this_char);
2117                         return 1;
2118                 }
2119         }
2120         return 0;
2121
2122 bad_val:
2123         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2124                value, this_char);
2125         return 1;
2126
2127 }
2128
2129 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2130 {
2131         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2132         unsigned long max_blocks = sbinfo->max_blocks;
2133         unsigned long max_inodes = sbinfo->max_inodes;
2134         int policy = sbinfo->policy;
2135         nodemask_t policy_nodes = sbinfo->policy_nodes;
2136         unsigned long blocks;
2137         unsigned long inodes;
2138         int error = -EINVAL;
2139
2140         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2141                                 &max_inodes, &policy, &policy_nodes))
2142                 return error;
2143
2144         spin_lock(&sbinfo->stat_lock);
2145         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2146         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2147         if (max_blocks < blocks)
2148                 goto out;
2149         if (max_inodes < inodes)
2150                 goto out;
2151         /*
2152          * Those tests also disallow limited->unlimited while any are in
2153          * use, so i_blocks will always be zero when max_blocks is zero;
2154          * but we must separately disallow unlimited->limited, because
2155          * in that case we have no record of how much is already in use.
2156          */
2157         if (max_blocks && !sbinfo->max_blocks)
2158                 goto out;
2159         if (max_inodes && !sbinfo->max_inodes)
2160                 goto out;
2161
2162         error = 0;
2163         sbinfo->max_blocks  = max_blocks;
2164         sbinfo->free_blocks = max_blocks - blocks;
2165         sbinfo->max_inodes  = max_inodes;
2166         sbinfo->free_inodes = max_inodes - inodes;
2167         sbinfo->policy = policy;
2168         sbinfo->policy_nodes = policy_nodes;
2169 out:
2170         spin_unlock(&sbinfo->stat_lock);
2171         return error;
2172 }
2173 #endif
2174
2175 static void shmem_put_super(struct super_block *sb)
2176 {
2177         kfree(sb->s_fs_info);
2178         sb->s_fs_info = NULL;
2179 }
2180
2181 static int shmem_fill_super(struct super_block *sb,
2182                             void *data, int silent)
2183 {
2184         struct inode *inode;
2185         struct dentry *root;
2186         int mode   = S_IRWXUGO | S_ISVTX;
2187         uid_t uid = current->fsuid;
2188         gid_t gid = current->fsgid;
2189         int err = -ENOMEM;
2190         struct shmem_sb_info *sbinfo;
2191         unsigned long blocks = 0;
2192         unsigned long inodes = 0;
2193         int policy = MPOL_DEFAULT;
2194         nodemask_t policy_nodes = node_online_map;
2195
2196 #ifdef CONFIG_TMPFS
2197         /*
2198          * Per default we only allow half of the physical ram per
2199          * tmpfs instance, limiting inodes to one per page of lowmem;
2200          * but the internal instance is left unlimited.
2201          */
2202         if (!(sb->s_flags & MS_NOUSER)) {
2203                 blocks = totalram_pages / 2;
2204                 inodes = totalram_pages - totalhigh_pages;
2205                 if (inodes > blocks)
2206                         inodes = blocks;
2207                 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2208                                         &inodes, &policy, &policy_nodes))
2209                         return -EINVAL;
2210         }
2211         sb->s_export_op = &shmem_export_ops;
2212 #else
2213         sb->s_flags |= MS_NOUSER;
2214 #endif
2215
2216         /* Round up to L1_CACHE_BYTES to resist false sharing */
2217         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2218                                 L1_CACHE_BYTES), GFP_KERNEL);
2219         if (!sbinfo)
2220                 return -ENOMEM;
2221
2222         spin_lock_init(&sbinfo->stat_lock);
2223         sbinfo->max_blocks = blocks;
2224         sbinfo->free_blocks = blocks;
2225         sbinfo->max_inodes = inodes;
2226         sbinfo->free_inodes = inodes;
2227         sbinfo->policy = policy;
2228         sbinfo->policy_nodes = policy_nodes;
2229
2230         sb->s_fs_info = sbinfo;
2231         sb->s_maxbytes = SHMEM_MAX_BYTES;
2232         sb->s_blocksize = PAGE_CACHE_SIZE;
2233         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2234         sb->s_magic = TMPFS_MAGIC;
2235         sb->s_op = &shmem_ops;
2236         sb->s_time_gran = 1;
2237 #ifdef CONFIG_TMPFS_POSIX_ACL
2238         sb->s_xattr = shmem_xattr_handlers;
2239         sb->s_flags |= MS_POSIXACL;
2240 #endif
2241
2242         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2243         if (!inode)
2244                 goto failed;
2245         inode->i_uid = uid;
2246         inode->i_gid = gid;
2247         root = d_alloc_root(inode);
2248         if (!root)
2249                 goto failed_iput;
2250         sb->s_root = root;
2251         return 0;
2252
2253 failed_iput:
2254         iput(inode);
2255 failed:
2256         shmem_put_super(sb);
2257         return err;
2258 }
2259
2260 static struct kmem_cache *shmem_inode_cachep;
2261
2262 static struct inode *shmem_alloc_inode(struct super_block *sb)
2263 {
2264         struct shmem_inode_info *p;
2265         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2266         if (!p)
2267                 return NULL;
2268         return &p->vfs_inode;
2269 }
2270
2271 static void shmem_destroy_inode(struct inode *inode)
2272 {
2273         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2274                 /* only struct inode is valid if it's an inline symlink */
2275                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2276         }
2277         shmem_acl_destroy_inode(inode);
2278         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2279 }
2280
2281 static void init_once(void *foo, struct kmem_cache *cachep,
2282                       unsigned long flags)
2283 {
2284         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2285
2286         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2287             SLAB_CTOR_CONSTRUCTOR) {
2288                 inode_init_once(&p->vfs_inode);
2289 #ifdef CONFIG_TMPFS_POSIX_ACL
2290                 p->i_acl = NULL;
2291                 p->i_default_acl = NULL;
2292 #endif
2293         }
2294 }
2295
2296 static int init_inodecache(void)
2297 {
2298         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2299                                 sizeof(struct shmem_inode_info),
2300                                 0, 0, init_once, NULL);
2301         if (shmem_inode_cachep == NULL)
2302                 return -ENOMEM;
2303         return 0;
2304 }
2305
2306 static void destroy_inodecache(void)
2307 {
2308         kmem_cache_destroy(shmem_inode_cachep);
2309 }
2310
2311 static const struct address_space_operations shmem_aops = {
2312         .writepage      = shmem_writepage,
2313         .set_page_dirty = __set_page_dirty_nobuffers,
2314 #ifdef CONFIG_TMPFS
2315         .prepare_write  = shmem_prepare_write,
2316         .commit_write   = simple_commit_write,
2317 #endif
2318         .migratepage    = migrate_page,
2319 };
2320
2321 static struct file_operations shmem_file_operations = {
2322         .mmap           = shmem_mmap,
2323 #ifdef CONFIG_TMPFS
2324         .llseek         = generic_file_llseek,
2325         .read           = shmem_file_read,
2326         .write          = shmem_file_write,
2327         .fsync          = simple_sync_file,
2328         .sendfile       = shmem_file_sendfile,
2329 #endif
2330 };
2331
2332 static struct inode_operations shmem_inode_operations = {
2333         .truncate       = shmem_truncate,
2334         .setattr        = shmem_notify_change,
2335         .truncate_range = shmem_truncate_range,
2336 #ifdef CONFIG_TMPFS_POSIX_ACL
2337         .setxattr       = generic_setxattr,
2338         .getxattr       = generic_getxattr,
2339         .listxattr      = generic_listxattr,
2340         .removexattr    = generic_removexattr,
2341         .permission     = shmem_permission,
2342 #endif
2343
2344 };
2345
2346 static struct inode_operations shmem_dir_inode_operations = {
2347 #ifdef CONFIG_TMPFS
2348         .create         = shmem_create,
2349         .lookup         = simple_lookup,
2350         .link           = shmem_link,
2351         .unlink         = shmem_unlink,
2352         .symlink        = shmem_symlink,
2353         .mkdir          = shmem_mkdir,
2354         .rmdir          = shmem_rmdir,
2355         .mknod          = shmem_mknod,
2356         .rename         = shmem_rename,
2357 #endif
2358 #ifdef CONFIG_TMPFS_POSIX_ACL
2359         .setattr        = shmem_notify_change,
2360         .setxattr       = generic_setxattr,
2361         .getxattr       = generic_getxattr,
2362         .listxattr      = generic_listxattr,
2363         .removexattr    = generic_removexattr,
2364         .permission     = shmem_permission,
2365 #endif
2366 };
2367
2368 static struct inode_operations shmem_special_inode_operations = {
2369 #ifdef CONFIG_TMPFS_POSIX_ACL
2370         .setattr        = shmem_notify_change,
2371         .setxattr       = generic_setxattr,
2372         .getxattr       = generic_getxattr,
2373         .listxattr      = generic_listxattr,
2374         .removexattr    = generic_removexattr,
2375         .permission     = shmem_permission,
2376 #endif
2377 };
2378
2379 static struct super_operations shmem_ops = {
2380         .alloc_inode    = shmem_alloc_inode,
2381         .destroy_inode  = shmem_destroy_inode,
2382 #ifdef CONFIG_TMPFS
2383         .statfs         = shmem_statfs,
2384         .remount_fs     = shmem_remount_fs,
2385 #endif
2386         .delete_inode   = shmem_delete_inode,
2387         .drop_inode     = generic_delete_inode,
2388         .put_super      = shmem_put_super,
2389 };
2390
2391 static struct vm_operations_struct shmem_vm_ops = {
2392         .nopage         = shmem_nopage,
2393         .populate       = shmem_populate,
2394 #ifdef CONFIG_NUMA
2395         .set_policy     = shmem_set_policy,
2396         .get_policy     = shmem_get_policy,
2397 #endif
2398 };
2399
2400
2401 static int shmem_get_sb(struct file_system_type *fs_type,
2402         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2403 {
2404         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2405 }
2406
2407 static struct file_system_type tmpfs_fs_type = {
2408         .owner          = THIS_MODULE,
2409         .name           = "tmpfs",
2410         .get_sb         = shmem_get_sb,
2411         .kill_sb        = kill_litter_super,
2412 };
2413 static struct vfsmount *shm_mnt;
2414
2415 static int __init init_tmpfs(void)
2416 {
2417         int error;
2418
2419         error = init_inodecache();
2420         if (error)
2421                 goto out3;
2422
2423         error = register_filesystem(&tmpfs_fs_type);
2424         if (error) {
2425                 printk(KERN_ERR "Could not register tmpfs\n");
2426                 goto out2;
2427         }
2428
2429         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2430                                 tmpfs_fs_type.name, NULL);
2431         if (IS_ERR(shm_mnt)) {
2432                 error = PTR_ERR(shm_mnt);
2433                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2434                 goto out1;
2435         }
2436         return 0;
2437
2438 out1:
2439         unregister_filesystem(&tmpfs_fs_type);
2440 out2:
2441         destroy_inodecache();
2442 out3:
2443         shm_mnt = ERR_PTR(error);
2444         return error;
2445 }
2446 module_init(init_tmpfs)
2447
2448 /*
2449  * shmem_file_setup - get an unlinked file living in tmpfs
2450  *
2451  * @name: name for dentry (to be seen in /proc/<pid>/maps
2452  * @size: size to be set for the file
2453  *
2454  */
2455 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2456 {
2457         int error;
2458         struct file *file;
2459         struct inode *inode;
2460         struct dentry *dentry, *root;
2461         struct qstr this;
2462
2463         if (IS_ERR(shm_mnt))
2464                 return (void *)shm_mnt;
2465
2466         if (size < 0 || size > SHMEM_MAX_BYTES)
2467                 return ERR_PTR(-EINVAL);
2468
2469         if (shmem_acct_size(flags, size))
2470                 return ERR_PTR(-ENOMEM);
2471
2472         error = -ENOMEM;
2473         this.name = name;
2474         this.len = strlen(name);
2475         this.hash = 0; /* will go */
2476         root = shm_mnt->mnt_root;
2477         dentry = d_alloc(root, &this);
2478         if (!dentry)
2479                 goto put_memory;
2480
2481         error = -ENFILE;
2482         file = get_empty_filp();
2483         if (!file)
2484                 goto put_dentry;
2485
2486         error = -ENOSPC;
2487         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2488         if (!inode)
2489                 goto close_file;
2490
2491         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2492         d_instantiate(dentry, inode);
2493         inode->i_size = size;
2494         inode->i_nlink = 0;     /* It is unlinked */
2495         file->f_vfsmnt = mntget(shm_mnt);
2496         file->f_dentry = dentry;
2497         file->f_mapping = inode->i_mapping;
2498         file->f_op = &shmem_file_operations;
2499         file->f_mode = FMODE_WRITE | FMODE_READ;
2500         return file;
2501
2502 close_file:
2503         put_filp(file);
2504 put_dentry:
2505         dput(dentry);
2506 put_memory:
2507         shmem_unacct_size(flags, size);
2508         return ERR_PTR(error);
2509 }
2510
2511 /*
2512  * shmem_zero_setup - setup a shared anonymous mapping
2513  *
2514  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2515  */
2516 int shmem_zero_setup(struct vm_area_struct *vma)
2517 {
2518         struct file *file;
2519         loff_t size = vma->vm_end - vma->vm_start;
2520
2521         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2522         if (IS_ERR(file))
2523                 return PTR_ERR(file);
2524
2525         if (vma->vm_file)
2526                 fput(vma->vm_file);
2527         vma->vm_file = file;
2528         vma->vm_ops = &shmem_vm_ops;
2529         return 0;
2530 }